scBench-Long
A verifiable benchmark for long-horizon single-cell biology in which agents must recover scientific conclusions from raw or near-raw data without prescribed methods. 21 evaluations spanning melanoma CD8 T-cell reactivity, CD8 RNA+ATAC regulatory inference, human-monkey chimera development, KRAS-driven lung tumor aging, and lethal COVID-19 lung pathology, with deterministic grading and trajectory rubrics over 1,068 completed agent runs.
Composite
63.5
Experimental validation
N/A — retrospective verifiable benchmark over published single-cell datasets
Stages
Virtual CellDisease modeling
Modalities
small molecule
Task types
classificationretrievalgeneration
Size
evaluations: 21
agent_runs: 1,068
splits: {'train': 0, 'val': 0, 'test': 21}
note: 21 deterministically-graded long-horizon evaluations across paired scRNA/TCR, RNA+ATAC, cross-species transcriptomics, combinatorial scRNA-seq, snRNA-seq and immune repertoires; 1,068 completed agent runs analyzed
agent_runs: 1,068
splits: {'train': 0, 'val': 0, 'test': 21}
note: 21 deterministically-graded long-horizon evaluations across paired scRNA/TCR, RNA+ATAC, cross-species transcriptomics, combinatorial scRNA-seq, snRNA-seq and immune repertoires; 1,068 completed agent runs analyzed
License
Other — see arXiv 2606.26563 (data derived from public single-cell studies)
First release
2026-06-25
Last updated
2026-06-25
Official site
Leaderboard
→ leaderboard
Dataset
Code / GitHub
→ repository
HuggingFace
→ HF
Paper
scBench-Long: Verifiable Benchmarking of Long-Horizon Single-Cell Biology · Ian Diks, Zhen Yang, Arjun Banerjee, Tim Proctor, Kenny Workman · 2026 · paper · doi:N/A — arXiv preprint 2606.26563 · 0 citations
Flags
none
Experts
Groups
Hosted by
—
Related benchmarks
Rubric (7-criterion)
rigor
4
coverage
4
maintenance
3
adoption
1
quality
4
accessibility
3
industry_relevance
3
Notes
Strong addition to the verifiable agentic-science genre: long-horizon tasks demand recovering biological conclusions from raw data without prescribed methods, with deterministic grading and trajectory rubrics (rigor 4, quality 4). Broad assay coverage across 21 evaluations including a KRAS-driven lung tumor aging task relevant to longevity/oncology (coverage 4). Adoption 1 as a June 2026 preprint; maintenance/accessibility pending public harness and leaderboard. Same authorship lineage as TxBench-PP.